Unmanned vehicle path planning method and system based on big data

Abstract

The invention belongs to the technical field of unmanned vehicles, and discloses an unmanned vehicle path planning method and system based on big data, and the method comprises the steps: obtaining traffic static data, traffic dynamic data and vehicle state data, and sequentially carrying out the data preprocessing and data fusion operation, and obtaining a vehicle traffic information data set; obtaining a vehicle path target and user preference, and performing path decision by using a pre-configured path decision model in combination with the vehicle traffic information data set to obtain a global initial path; and continuously acquiring real-time data of the vehicle in the process of executing the global initial path, and performing dynamic path adjustment by using a dynamic obstacle avoidance algorithm and a comprehensive cost function until the vehicle arrives at a destination. According to the method, the safety and efficiency of the unmanned vehicle are remarkably improved, the method is closer to the cognition and decision logic of real human driving, and the method has high engineering application value and popularization prospects.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of unmanned vehicles, in particular to an unmanned vehicle path planning method and system based on big data. BACKGROUND

[0002] The future application prospect of unmanned vehicles is broad, and is expected to trigger revolutionary changes in many fields. Its core value has gradually emerged in the dimensions of safety, efficiency, cost and special scene adaptation. In terms of traffic safety and operation efficiency, unmanned technology can greatly avoid traffic accident risks caused by human operation errors, and optimize road traffic efficiency through intelligent vehicle flow regulation, relying on accurate path planning algorithms and real-time dynamic data feedback.

[0003] In the field of people's livelihood and industrial services, the landing of unmanned technology will reshape the travel and logistics pattern: in the shared travel scenario, it can reduce the cost of manual driving, and improve travel convenience through 24-hour uninterrupted service and dynamic route adjustment; in the logistics transportation scenario, it can optimize the "door-to-door" transportation efficiency and promote the transformation of traditional freight mode to intelligent and unmanned, relying on stable driving strategy and real-time monitoring of cargo status. In special operating environments such as mining areas, agriculture and industrial parks, unmanned vehicles can not only adapt to rugged terrain, dust pollution and high-intensity operations in harsh conditions, but also improve operation accuracy and safety factor through standardized path planning, and reduce the risk of personnel casualties.

[0004] Although the potential is huge, the current unmanned path planning technology still has obvious application limitations. In the civilian manned scenario, although the existing scheme can generate a global path, it is difficult to balance "obstacle avoidance safety" and "passenger comfort" when facing complex road conditions such as "pedestrians crossing, vehicles queuing, temporary construction", and the adaptability to users' individualized needs such as "efficiency first", "low energy consumption first" and "smooth first" is insufficient, which is prone to problems such as "frequent lane changes to avoid obstacles" and "recommended route and user preferences are out of touch". In harsh road conditions or densely populated special operating scenarios, the technical shortcomings are also prominent: on the one hand, the integration capability of scene data such as "mining area steep and rocky road, park peak flow" is weak, and the environmental perception is not comprehensive enough; on the other hand, there is a lack of dynamic adjustment mechanism of "safety first + efficient scheduling", and some schemes are prone to delay due to excessive emphasis on obstacle avoidance, or ignore potential risks in pursuit of efficiency, which is difficult to meet the fine operation needs in special scenarios. SUMMARY

[0005] The embodiment of the present application provides an unmanned vehicle path planning method and system based on big data to solve the problems in the prior art.

[0006] The following presents a simplified summary of some aspects of the disclosure in order to provide a basic understanding of such aspects. This summary is not an extensive overview of the disclosure and is not intended to identify key / critical elements or to delineate the scope of the disclosure. Its sole purpose is to present some concepts in a simplified form as a prelude to the more detailed description that is presented later.

[0007] According to a first aspect of embodiments of the present application, a big data based path planning method for an unmanned vehicle is provided.

[0008] In one embodiment, a big data based path planning method for an unmanned vehicle comprises:

[0009] Obtaining traffic static data, traffic dynamic data and vehicle state data, and sequentially performing data preprocessing and data fusion operations to obtain a vehicle traffic information dataset;

[0010] Obtaining a vehicle path target and user preference, and combining the vehicle traffic information dataset, using a pre-configured path decision model to make a path decision to obtain a global initial path;

[0011] Continuously obtaining real-time data of the vehicle in the process of executing the global initial path, and using a dynamic obstacle avoidance algorithm and a comprehensive cost function to make dynamic path adjustment until the destination is reached.

[0012] In one embodiment, obtaining traffic static data, traffic dynamic data and vehicle state data, and sequentially performing data preprocessing and data fusion operations to obtain a vehicle traffic information dataset comprises the following steps:

[0013] Based on a high-precision map database, a traffic management platform and a vehicle-mounted sensor, obtaining traffic static data, traffic dynamic data and vehicle state data;

[0014] Respectively performing cleaning, spatio-temporal alignment and standardization processing on the traffic static data, the traffic dynamic data and the vehicle state data to obtain pre-processed traffic static data, traffic dynamic data and vehicle state data;

[0015] Using a heterogeneous data fusion algorithm, performing semantic association and feature integration on the pre-processed traffic static data, traffic dynamic data and vehicle state data to obtain a vehicle traffic information dataset.

[0016] In one embodiment, using a heterogeneous data fusion algorithm, performing semantic association and feature integration on the pre-processed traffic static data, traffic dynamic data and vehicle state data to obtain a vehicle traffic information dataset comprises the following steps:

[0017] establish semantic association among the traffic static data, the traffic dynamic data and the vehicle state data; and based on spatial information and time information, perform feature extraction on the preprocessed traffic static data, the traffic dynamic data and the vehicle state data to obtain traffic static features, traffic dynamic features and vehicle state features;

[0018] perform standardization processing on the traffic static features, the traffic dynamic features and the vehicle state features; and based on a weighted fusion strategy of confidence, fuse the traffic static features, the traffic dynamic features and the vehicle state features to obtain a set feature set;

[0019] based on path planning requirements, structurally organize the set feature set to generate a vehicle traffic information data set containing three dimensions of "road basic attributes-real-time traffic state-vehicle dynamic adaptability".

[0020] In one embodiment, the vehicle path target and the user preference are obtained, and combined with the vehicle traffic information data set, a pre-configured path decision model is used for path decision to obtain a global initial path, including the following steps:

[0021] obtain the path target and the user preference;

[0022] based on the path target, filter the vehicle traffic information data set to obtain filtered vehicle traffic information;

[0023] convert the user preference into path planning constraint conditions;

[0024] based on an improved A-star algorithm and a reinforcement learning strategy, train a path decision model;

[0025] use the path decision model, and combine the filtered vehicle traffic information and the path planning constraint conditions to perform path search to obtain the global initial path.

[0026] In one embodiment, based on the improved A-star algorithm and the reinforcement learning strategy, training the path decision model includes the following steps:

[0027] based on a pre-constructed historical data set, combine the improved A-star algorithm to obtain optimal path samples;

[0028] construct a reinforcement learning model; and according to the optimal path samples, train the reinforcement learning model to obtain an initial path decision model;

[0029] use the historical data set to iteratively train the initial path decision model, optimize the hyperparameters of the initial path decision model through cross-validation, and obtain the path decision model.

[0030] In one embodiment, a reinforcement learning model is constructed; and the reinforcement learning model is trained according to the optimal path samples to obtain an initial path decision model, including the following steps:

[0031] The state space, action space and reward function of the reinforcement learning model are defined;

[0032] A neural network structure of the reinforcement learning model including an input layer, a hidden layer and an output layer is constructed;

[0033] The constructed reinforcement learning model is trained based on the optimal path samples to obtain the initial path decision model.

[0034] In one embodiment, the expression formula of the reward function is as follows:

[0035]

[0036] In the formula, R(s, a, s') represents an immediate reward value obtained when the action a is performed under the state s to transfer to the new state s'; ω1, ω2, ω3 and ω4 respectively represent weight coefficients of the travel time reward, the path distance reward, the sample similarity reward and the terminal reward; R time (s, a, s') represents the travel time reward; T(s, a, s') represents the actual travel time from the state s to the state s' by performing the action a; T ∗ (s, s goal ) represents the optimal travel time from the state s to the target state s goal calculated by the improved A* algorithm; α represents a time penalty coefficient, controlling the influence degree of the time deviation on the reward; R dist (s, a, s') represents the path distance reward; D(s, a, s') represents the actual path distance from the state s to the state s' by performing the action a; D ∗ (s, s goal ) represents the optimal path distance from the state s to the target state s goal calculated by the improved A* algorithm; β represents a distance penalty coefficient, controlling the influence degree of the distance deviation on the reward; R sim (s, a, s') represents the sample similarity reward; a * (s) represents the expert recommended action under the state s in the optimal path sample; I(·) represents an indicator function; γ represents a similarity reward coefficient, encouraging the model to imitate the decision logic of the optimal sample; R term (s, a, s') represents the terminal reward; s goal represents the target state of the path planning; δ represents a terminal reward coefficient.

[0037] In one embodiment, real-time data of the vehicle in performing the global initial path process is continuously acquired, and dynamic path adjustment is performed by using a dynamic obstacle avoidance algorithm and a comprehensive cost function until the destination is reached, including the following steps:

[0038] Continuously acquiring real-time data of the vehicle in performing the global initial path process;

[0039] Fusing and updating the real-time data with the vehicle traffic information dataset to build a local dynamic environment model at the current time;

[0040] Based on the global initial path and in combination with the local dynamic environment model, path adjustment is performed by using a dynamic obstacle avoidance algorithm and a comprehensive cost function to obtain an adjusted path and execute the same;

[0041] Continuously acquiring real-time data in performing the adjusted path, updating the local dynamic environment model, and repeatedly performing path adjustment according to the updated local dynamic environment model in combination with the dynamic obstacle avoidance algorithm and the comprehensive cost function;

[0042] Real-time comparison of the current position of the vehicle with the destination coordinates, and when the distance between the two is less than a preset threshold, the destination is determined to be reached, and the path planning and adjustment process is ended.

[0043] In one embodiment, based on the global initial path and in combination with the local dynamic environment model, path adjustment is performed by using a dynamic obstacle avoidance algorithm to obtain an adjusted path and execute the same, including the following steps:

[0044] Based on the global initial path and in combination with the local dynamic environment model, a plurality of candidate paths are obtained according to the dynamic obstacle avoidance algorithm;

[0045] A comprehensive cost function including safety, comfort and path tracking is constructed, and each candidate path is scored based on the comprehensive cost function;

[0046] The candidate path with the minimum comprehensive cost is selected as the optimal local path, and the optimal local path is verified for smoothness and feasibility;

[0047] The verified optimal local path is taken as the adjusted path and executed.

[0048] According to a second aspect of the embodiment of the present application, a big data-based path planning system for an unmanned vehicle is provided.

[0049] In one embodiment, a big data-based path planning system for an unmanned vehicle includes:

[0050] A vehicle traffic information dataset construction module is configured to acquire traffic static data, traffic dynamic data and vehicle state data, and sequentially perform data preprocessing and data fusion operations to obtain a vehicle traffic information dataset;

[0051] a global initial path construction module configured to obtain a vehicle path target and a user preference, and combine a vehicle traffic information dataset to perform path decision by using a pre-configured path decision model to obtain a global initial path;

[0052] a real-time local path adjustment module configured to continuously obtain real-time data of the vehicle in the process of executing the global initial path, and perform dynamic path adjustment by using a dynamic obstacle avoidance algorithm and a comprehensive cost function until the destination is reached.

[0053] According to a third aspect of the embodiments of the present application, a computer device is provided.

[0054] In some embodiments, the computer device comprises a memory and a processor, the memory stores a computer program, and the processor implements the steps of the above method when executing the computer program.

[0055] According to a fourth aspect of the embodiments of the present application, a computer readable storage medium is provided.

[0056] In one embodiment, the computer readable storage medium stores a computer program, and the computer program is executed by a processor to implement the steps of the above method.

[0057] The technical solutions provided by the embodiments of the present application can include the following beneficial effects:

[0058] 1. The method realizes deep cognition of complex traffic environment through big data fusion, generates a global path considering user intention by using a path decision model, and ensures local safety by combining a real-time dynamic obstacle avoidance algorithm, thereby significantly improving the safety and efficiency of the unmanned vehicle, and being closer to the cognition and decision logic of real human driving, and having strong engineering application value and promotion prospect.

[0059] 2. The method constructs a comprehensive environment perception model through multi-source data fusion, combines constraint conditions and a comprehensive cost function converted from user preference, can realize accurate balance between safety and comfort in civilian manned scenarios, meet individual travel needs of different users, can integrate special scenario exclusive data, adapt to special environments such as mining areas and park areas by using a dynamic adjustment strategy of "safety first and efficiency second", and at the same time, relies on a whole-process closed-loop design of "real-time data acquisition-model updating-path adjustment" to ensure that local adjustment does not deviate from the global target, and greatly improves the adaptability and execution stability of the path in complex environments.

[0060] It should be understood that the foregoing general description and the following detailed description are only exemplary and explanatory, and cannot limit the present application. BRIEF DESCRIPTION OF DRAWINGS

[0061] The accompanying drawings, which are incorporated herein and constitute part of the specification, illustrate embodiments consistent with the present application and, together with the description, further serve to explain the principles of the application.

[0062] Figure 1 is a flow chart of a big data based path planning method for unmanned vehicle according to an exemplary embodiment;

[0063] Figure 2 is a principle block diagram of a big data based path planning system for unmanned vehicle according to an exemplary embodiment;

[0064] Figure 3 is a structural schematic diagram of a computer device according to an exemplary embodiment. DETAILED DESCRIPTION

[0065] The following description and drawings are illustrative of specific embodiments of the present document and are not intended to limit the scope of the present document. Parts and features of some embodiments can be included or substituted in other embodiments. The scope of the embodiments of the present document encompasses the whole scope of the claims and all available equivalents of the claims. In this document, the terms "first", "second" and the like are used to distinguish one element from another, but do not require or imply that these elements have to be in a certain actual relationship or order. In fact, the first element can also be referred to as the second element, and vice versa. Moreover, the terms "comprise", "include" or any other variant thereof are intended to cover non-exclusive inclusion, so that a structure, device or apparatus that comprises a list of elements does not only include those elements, but also includes other elements not expressly listed or inherent to such structure, device or apparatus. Without more limitations, an element defined by the phrase "comprising a" does not exclude the presence of additional identical elements in the structure, device or apparatus comprising the element. Various embodiments are described in a progressive manner, each focusing on the differences from other embodiments, and the same or similar parts between various embodiments can be referred to each other.

[0066] In this document, the term "plurality" means two or more, unless otherwise specified.

[0067] In this document, the character " / " represents an "or" relationship between the objects before and after it. For example, A / B means A or B.

[0068] In this document, the term "and / or" is a description of the relationship between the objects, which means that there can be three relationships. For example, A and / or B means that there are three relationships of A or B, or A and B.

[0069] It should be understood that although the steps in the flowchart are shown in sequential order, such that each step depends on completion of the previous step before execution of the next step, the steps are not necessarily performed in the order shown by the arrows. Unless specifically stated otherwise, the steps can be performed in other orders. Moreover, at least some of the steps can include multiple sub-steps or stages, which are not necessarily performed at the same time, but can be performed at different times, and which can be performed in a different order than shown, in parallel with, or alternating with, at least some of the other steps or sub-steps or stages of other steps.

[0070] Each module in the device or system of the present application can be implemented wholly or partially by software, hardware, and combinations thereof. The modules can be embedded in or independent of a processor in a computer device in hardware form, or stored in a memory in a computer device in software form, so as to be called and executed by a processor to perform operations corresponding to the modules.

[0071] In the case of no conflict, the embodiments in the present application and the features in the embodiments can be combined with each other.

[0072] Figure 1 An embodiment of a big data-based path planning method for an unmanned vehicle is shown.

[0073] In this optional embodiment, the big data-based path planning method for an unmanned vehicle comprises:

[0074] S101, traffic static data, traffic dynamic data, and vehicle state data are acquired, and data preprocessing and data fusion operations are sequentially performed to obtain a vehicle traffic information dataset.

[0075] It should be explained that by sequentially performing data acquisition, cleaning, time synchronization, coordinate unification, standardization, and multi-level fusion, heterogeneous, heterogeneous, and asynchronous original traffic static data, traffic dynamic data, and vehicle state data are converted into vehicle traffic information datasets with consistent structure, rich semantics, and spatiotemporal alignment. The vehicle traffic information dataset is used as an input of a path decision model, fully depicts the static environment, dynamic traffic state, and self-capability of the vehicle, and provides a reliable data basis for intelligent decision-making.

[0076] In this optional embodiment, the acquisition of traffic static data, traffic dynamic data, and vehicle state data, and the sequential performance of data preprocessing and data fusion operations to obtain a vehicle traffic information dataset comprises the following steps:

[0077] Based on a high-precision map database, a traffic management platform, and a vehicle-mounted sensor, traffic static data, traffic dynamic data, and vehicle state data are obtained.

[0078] It should be explained that the traffic static data can be collected by the high-precision map database to obtain high-definition map information, including road structure, traffic signs, signal lamp position, etc. The traffic dynamic data can be obtained by the traffic management platform and V2X communication (vehicle networking communication) to obtain real-time traffic flow information, such as traffic congestion, accident reports, and road construction information. The vehicle state data can be obtained by the vehicle-mounted sensor (such as GPS, laser radar, camera, etc.) to obtain the vehicle's own state and surrounding environment information.

[0079] The traffic static data, traffic dynamic data, and vehicle state data are respectively cleaned, time-space aligned, and standardized to obtain pre-processed traffic static data, traffic dynamic data, and vehicle state data.

[0080] It should be explained that the original traffic static data, traffic dynamic data, and vehicle state data are respectively subjected to denoising and outlier rejection data cleaning operations. In terms of time, due to different sampling frequencies of various data, it is necessary to unify to the same time reference, and linear interpolation or spline interpolation is used for alignment of non-synchronous data. In terms of space, all spatial data are converted to a local coordinate system centered on the ego vehicle. Finally, in order to eliminate dimensional differences, numerical value type variables are standardized for subsequent fusion.

[0081] The pre-processed traffic static data, traffic dynamic data, and vehicle state data are subjected to semantic association and feature integration by using a heterogeneous data fusion algorithm to obtain a vehicle traffic information dataset.

[0082] In this optional embodiment, the pre-processed traffic static data, traffic dynamic data, and vehicle state data are subjected to semantic association and feature integration by using a heterogeneous data fusion algorithm to obtain a vehicle traffic information dataset, including the following steps:

[0083] The semantic association between the traffic static data, traffic dynamic data, and vehicle state data is established, and the pre-processed traffic static data, traffic dynamic data, and vehicle state data are subjected to feature extraction based on spatial information and time information to obtain traffic static features, traffic dynamic features, and vehicle state features.

[0084] The traffic static features, traffic dynamic features, and vehicle state features are subjected to standardization processing, and the traffic static features, traffic dynamic features, and vehicle state features are fused based on a confidence weighted fusion strategy to obtain a set feature set.

[0085] Based on the path planning requirements, the set feature set is structured and organized to generate a vehicle traffic information dataset containing three dimensions of "road basic attributes-real-time traffic state-vehicle dynamic adaptability".

[0086] It should be explained that the steps of the above-mentioned fusion vehicle traffic information dataset method are described as follows in combination with specific implementation scenarios:

[0087] Step 1, semantic association and feature extraction.

[0088] Taking "vehicle A passing through XX section at 8:30" as the association core, the three types of data are bound. Among them, the traffic static data (the speed limit of XX section is 60 km / h, 3 lanes, and the road material is asphalt); traffic dynamic data (the real-time traffic flow of the section at 8:30 is 200 vehicles / hour, the average speed is 45 km / h, and the sudden accident warning); vehicle state data (the position coordinates of vehicle A at 8:30, instantaneous speed 50 km / h, remaining power 60%). The semantic mapping relationship of the three is established through the time stamp (8:30) and the spatial coordinates (the latitude and longitude range of XX section).

[0089] Feature extraction is performed on the preprocessed traffic static data, traffic dynamic data and vehicle state data. Traffic static features such as "road speed limit value (60 km / h), number of lanes (3), road grade (main road), intersection spacing (500 m)" are extracted from traffic static data; traffic dynamic features such as "real-time flow rate (200 vehicles / hour), speed fluctuation value (±10 km / h), congestion index (0.6, 0-1 interval), accident influence range (100 m)" are extracted from traffic dynamic data; vehicle state features such as "instantaneous speed (50 km / h), acceleration (1.2 m / s²), remaining endurance (120 km), vehicle size (length 5 m / width 2 m)" are extracted from vehicle state data.

[0090] Step 2, standardization and weighted fusion.

[0091] The traffic static features, traffic dynamic features and vehicle state features are standardized. Numerical features are standardized using min-max normalization, such as "speed limit 60 km / h" and "real-time speed 45 km / h" are uniformly mapped to the [0, 1] interval (60 to 1.0, 45 to 0.75); Categorical features are converted into one-hot encoding, such as "road grade (main road / secondary road)" is converted into [1, 0] or [0, 1] vector; Boolean features (such as "accident warning") are converted into 0 / 1 values (1 for warning, 0 for no).

[0092] The traffic static features, traffic dynamic features and vehicle state features are fused based on a confidence-based weighted fusion strategy. Confidence weights are set, the traffic static data (sensor error < 5%) has high confidence (weight 0.4), the traffic dynamic data (high real-time but large fluctuation) has weight 0.3, and the vehicle state data (high-frequency acquisition is reliable) has weight 0.3.

[0093] The fusion calculation of “XX section-8:30” is: road traffic capacity feature = 0.4 x (lane number standardized value) + 0.3 x (vehicle flow standardized value) + 0.3 x (vehicle size adaptation value). Finally, a set of feature sets containing 12 fusion features such as “road traffic capacity, real-time risk coefficient, vehicle adaptation degree” is generated.

[0094] Step 3, structured organization generates data set.

[0095] Based on the path planning requirements (such as “optimal path calculation from start to end”), the set of feature sets is structured in three dimensions. The road basic attribute dimension includes static features such as “speed limit, lane number, road grade, intersection type”; the real-time traffic state dimension includes dynamic features such as “flow rate, congestion index, accident warning, signal light period”; and the vehicle dynamic adaptability dimension includes features such as “vehicle size and lane width matching degree, remaining battery capacity and road length adaptability, vehicle speed and speed limit deviation rate”, forming a vehicle traffic information data set containing “road basic attribute-real-time traffic state-vehicle dynamic adaptability” three dimensions.

[0096] The finally generated vehicle traffic information data set can directly support the path decision model. For example: the model can quickly filter the feasible path by querying “the real-time congestion index of XX section (dynamic dimension)” and “whether the vehicle remaining battery capacity can cover the section (adaptability dimension)”.

[0097] S102, obtain the vehicle path target and user preference, and combine the vehicle traffic information data set to use the pre-configured path decision model for path decision to obtain the global initial path.

[0098] It needs to be explained that this process realizes the global path planning of deeply fusing user intention and big data environment cognition.

[0099] In this optional embodiment, obtaining the vehicle path target and user preference, and combining the vehicle traffic information data set to use the pre-configured path decision model for path decision to obtain the global initial path includes the following steps:

[0100] Obtain the path target and user preference;

[0101] Based on the path target, the vehicle traffic information data set is filtered to obtain the filtered vehicle traffic information;

[0102] transforming the user preferences into path planning constraints;

[0103] training a path decision model based on the improved A-star algorithm and reinforcement learning strategy;

[0104] performing path search using the path decision model in combination with the filtered vehicle traffic information and path planning constraints to obtain a global initial path.

[0105] It should be explained that the multi-dimensional dynamic attributes and personalized preferences are transformed into an optimizable objective function using the "pre-configured path decision model", thereby generating a truly intelligent and adaptive global driving strategy.

[0106] In this optional embodiment, training a path decision model based on the improved A-star algorithm and reinforcement learning strategy includes the following steps:

[0107] obtaining optimal path samples based on the pre-constructed historical data set in combination with the improved A-star algorithm;

[0108] constructing a reinforcement learning model and training the reinforcement learning model based on the optimal path samples to obtain an initial path decision model;

[0109] iteratively training the initial path decision model using the historical data set, optimizing the hyperparameters of the initial path decision model through cross-validation, and obtaining the path decision model.

[0110] In this optional embodiment, constructing a reinforcement learning model and training the reinforcement learning model based on the optimal path samples to obtain an initial path decision model includes the following steps:

[0111] defining the state space, action space, and reward function of the reinforcement learning model;

[0112] constructing a neural network structure of the reinforcement learning model including an input layer, a hidden layer, and an output layer;

[0113] training the constructed reinforcement learning model based on the optimal path samples to obtain the initial path decision model.

[0114] It should be explained that training a path decision model based on the improved A-star algorithm and reinforcement learning strategy includes the following steps:

[0115] Step 1, define the core elements of the reinforcement learning model.

[0116] Define the state space (such as vehicle position, road congestion state, remaining distance, etc.), action space (such as turning, straight driving, etc.), and reward function;

[0117] The expression formula of the reward function is as follows:

[0118]

[0119] In the formula represents the instant reward value obtained when the action a under the state s is transferred to the new state s'; ω1, ω2, ω3, and ω4 respectively represent the weight coefficients of the travel time reward, the path distance reward, the sample similarity reward, and the terminal reward (ω1+ω2+ω3+ω4=1), which are used to adjust the importance of different factors (for example, ω1 can be increased when the user pays more attention to time); R time (s,a,s′) represents the travel time reward; T(s,a,s′) represents the actual travel time from the state s to the state s' after performing the action a (which can be calculated through real-time traffic data); T ∗ (s,s goal ) represents the optimal travel time from the state s to the target state s goal calculated by the improved A-star algorithm (the time in the optimal path sample); α represents the time penalty coefficient (α>0), which controls the influence degree of the time deviation on the reward; R dist (s,a,s′) represents the path distance reward; D(s,a,s′) represents the actual path distance from the state s to the state s' after performing the action a; D ∗ (s,s goal ) represents the optimal path distance from the state s to the target state s goal calculated by the improved A-star algorithm (the distance in the optimal path sample); β represents the distance penalty coefficient (β>0), which controls the influence degree of the distance deviation on the reward; R sim (s,a,s′) represents the sample similarity reward; a * (s) represents the expert recommended action (such as "straight" or "left turn") under the state s in the optimal path sample, which takes the value 1 when the condition in the bracket is true, and takes the value 0 otherwise (i.e., the model action is consistent with the expert action, and a positive reward is obtained, otherwise 0 is obtained); I(⋅) represents the indicator function; γ represents the similarity reward coefficient (γ>0), which encourages the model to imitate the decision logic of the optimal sample; R term (s,a,s′) represents the terminal reward; s goal represents the target state of path planning (such as the end point coordinates); δ represents the terminal reward coefficient (δ≫0), which provides significant positive reward for reaching the end point, and accelerates the convergence of the model.

[0120] Step 2, select the reinforcement learning algorithm framework.

[0121] Select a reinforcement learning algorithm framework, and build a neural network structure containing an input layer (receiving state features), a hidden layer (extracting features and making decision calculations), and an output layer (outputting action probabilities or Q values).

[0122] Step 3, preprocessing the optimal path samples.

[0123] Convert the path samples into sequence data (such as state-action-reward trajectory pairs) that can be recognized by reinforcement learning, and divide the training set and validation set.

[0124] Step 4, model training using imitation learning strategy.

[0125] Use the optimal path samples as expert demonstrations, and use behavior cloning (use state-action pairs in the sample directly for training, so that the model imitates the behavior of the expert) or inverse reinforcement learning (backpropagate the reward function from the sample and then train it) to make the model learn the path decision pattern.

[0126] When the average deviation of the decision results of the model on the validation set from the optimal path samples is below the preset threshold, or the number of training iterations reaches the set upper limit, optimize the hyperparameters of the initial path decision model through cross-validation to obtain the path decision model.

[0127] S103, continuously acquire real-time data of the vehicle during the execution of the global initial path, and use dynamic obstacle avoidance algorithm and comprehensive cost function for dynamic path adjustment until reaching the destination.

[0128] It needs to be explained that the global path is taken as a reference, combined with real-time dynamic environment, and an executable path that is both safe and comfortable is generated through optimization method to ensure the smooth operation of the unmanned vehicle in complex traffic.

[0129] In this optional embodiment, continuously acquiring real-time data of the vehicle during the execution of the global initial path, and using dynamic obstacle avoidance algorithm and comprehensive cost function for dynamic path adjustment until reaching the destination includes the following steps:

[0130] Continuously acquire real-time data of the vehicle during the execution of the global initial path;

[0131] Fuse and update the real-time data with the vehicle traffic information dataset to build a local dynamic environment model at the current time;

[0132] Based on the global initial path, combined with the local dynamic environment model, use dynamic obstacle avoidance algorithm and comprehensive cost function for path adjustment to obtain the adjusted path and execute it;

[0133] continuously acquire real-time data in the process of executing the adjusted path, update the local dynamic environment model, and repeatedly perform path adjustment according to the updated local dynamic environment model, the dynamic obstacle avoidance algorithm, and the comprehensive cost function;

[0134] The current position of the vehicle is compared with the destination coordinates in real time. When the distance between the two is less than a preset threshold, it is determined that the destination has been reached, and the path planning and adjustment process is ended.

[0135] It should be explained that through the vehicle-mounted sensors (cameras, millimeter wave radars, laser radars), GPS positioning systems, and vehicle-to-everything (V2X) modules, multi-dimensional data is collected in real time. A heterogeneous data fusion algorithm is used to perform spatio-temporal alignment of the real-time data and the pre-stored vehicle traffic information dataset, eliminate data conflicts, and define a local range of 500 meters to 1 kilometer with the current position of the vehicle as the center to construct a dynamic environment model.

[0136] The dynamic environment model includes a road topology sub-model, an obstacle sub-model, a traffic rule sub-model, and a global path reference sub-model. The road topology sub-model includes the number of available lanes, lane boundary coordinates, intersection positions, and turning rules. The obstacle sub-model includes the real-time position and future 3-5 second motion trajectory prediction of dynamic obstacles, and the no-entry area of static obstacles. The traffic rule sub-model includes the current road segment speed limit, signal light phase timing, and lane usage restrictions (such as bus lane restricted time period). The global path reference sub-model includes the road segment coordinates and estimated arrival time of the global initial path within the local range.

[0137] Based on the global initial path and in combination with the local dynamic environment model, the path is adjusted using the dynamic obstacle avoidance algorithm to obtain an adjusted path and execute it. Real-time data in the process of executing the adjusted path is continuously acquired to update the local dynamic environment model. According to the updated local dynamic environment model, the dynamic obstacle avoidance algorithm and the comprehensive cost function are combined to repeatedly perform path adjustment.

[0138] According to the scene setting threshold (such as 5 meters for urban roads and 1 meter for parking lot entrances), combined with the GPS positioning accuracy (±1 meter), it is determined that the path planning and adjustment process is ended after reaching the destination, and the path adjustment record of this trip is saved.

[0139] In this optional embodiment, based on the global initial path and in combination with the local dynamic environment model, the path is adjusted using the dynamic obstacle avoidance algorithm to obtain an adjusted path and execute it, which includes the following steps:

[0140] Based on the global initial path, in combination with the local dynamic environment model, and according to the dynamic obstacle avoidance algorithm, a plurality of candidate paths are obtained.

[0141] A comprehensive cost function including safety, comfort and path tracking is constructed, and each candidate path is scored based on the comprehensive cost function;

[0142] The candidate path with the minimum comprehensive cost is selected as the optimal local path, and the optimal local path is verified for smoothness and feasibility;

[0143] The verified optimal local path is taken as the adjusted path and is executed.

[0144] It needs to be explained that, compared with the road section of the global initial path in the local range and the local dynamic environment model, the conflict point (such as path overlapping with obstacles, lane closure leading to impassability) is identified; a dynamic obstacle avoidance algorithm is called to generate 3-5 candidate paths for avoiding the conflict under the constraint of the local dynamic environment model, each path including a coordinate sequence, an expected travel time and a turning node; a multi-objective cost function is constructed, and the expression formula of the multi-objective cost function is as follows:

[0145] C=ω 安全 C 安全 +ω 合规 C 合规 +ω 效率 C 效率 +ω 平滑 C 平滑 ;

[0146] In the formula, C 安全 represents the minimum distance to the obstacle (the smaller the distance, the higher the cost); C 合规 represents whether the traffic rules are violated (such as occupying a bus lane, running a red light, and the cost increases significantly when the rules are violated); C 效率 represents the time deviation from the global initial path (the greater the deviation, the higher the cost); C 平滑 represents the path curvature change rate (the more drastic the change, the higher the cost); ω 安全 , ω 合规 , ω 效率 , ω 平滑 all represent weight coefficients (the safety weight is the highest, such as ω 安全 =0.4).

[0147] The candidate path with the minimum comprehensive cost is selected, and whether it meets the vehicle dynamics constraints (such as minimum turning radius) and the continuity with the current driving state (such as speed matching) is verified, and the path is smoothed (such as Bessel curve optimization); path execution: the adjusted path is converted into vehicle control instructions (such as steering angle, throttle / brake opening) and executed.

[0148] Figure 2 An embodiment of an unmanned vehicle path planning system based on big data is shown.

[0149] In the optional embodiment, the unmanned vehicle path planning system based on big data comprises:

[0150] A vehicle traffic information dataset construction module 201 is configured to acquire traffic static data, traffic dynamic data and vehicle state data, and sequentially perform data preprocessing and data fusion to obtain a vehicle traffic information dataset;

[0151] A global initial path construction module 202 is configured to acquire a vehicle path target and user preference, and combine the vehicle traffic information dataset to perform path decision by using a pre-configured path decision model to obtain a global initial path;

[0152] A real-time local path adjustment module 203 is configured to continuously acquire real-time data of the vehicle in the process of executing the global initial path, and perform dynamic path adjustment by using a dynamic obstacle avoidance algorithm and a comprehensive cost function until the destination is reached.

[0153] In one embodiment, a computer device, which can be a server, has an internal structure as shown in Figure 3 The computer device comprises a processor, a memory and a network interface connected through a system bus. The processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device comprises a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium to run. The database of the computer device is configured to store static information and dynamic information data. The network interface of the computer device is configured to communicate with an external terminal through a network connection. The computer program is executed by the processor to implement the steps in the above method embodiments.

[0154] Those skilled in the art can understand that Figure 3 The structure shown in the above embodiment is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. Specifically, the computer device can comprise more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0155] In addition, the present application also provides a computer device comprising a memory and a processor. The memory stores a computer program. The processor implements the steps in the above method embodiments when executing the computer program.

[0156] In addition, the present application also provides a computer readable storage medium having a computer program stored thereon. The computer program is executed by the processor to implement the steps in the above method embodiments.

[0157] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium, and when executed, can include the processes of the above-mentioned embodiments of the methods. Any reference to memory, storage, database or other medium used in the embodiments provided by the present application can include at least one of non-volatile and volatile memory. The non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory or optical memory, etc. The volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, the RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0158] The present application is not limited to the structures described above and shown in the drawings, and various modifications and changes can be made without departing from the scope thereof. The scope of the present application is only limited by the appended claims.

Claims

1. A path planning method for autonomous vehicles based on big data, characterized in that, The method includes: Acquire static traffic data, dynamic traffic data, and vehicle status data, and perform data preprocessing and data fusion operations in sequence to obtain a vehicle traffic information dataset; The system obtains vehicle route goals and user preferences, and combines them with a vehicle traffic information dataset to make route decisions using a pre-configured route decision model, thereby obtaining a global initial route. The system continuously acquires real-time data of the vehicle during the execution of the global initial path and uses dynamic obstacle avoidance algorithms and comprehensive cost functions to dynamically adjust the path until the destination is reached.

2. The method for unmanned vehicle path planning based on big data according to claim 1, characterized in that, The process of acquiring static traffic data, dynamic traffic data, and vehicle status data, and then performing data preprocessing and data fusion operations in sequence to obtain a vehicle traffic information dataset includes the following steps: Based on high-precision map databases, traffic management platforms, and vehicle sensors, static traffic data, dynamic traffic data, and vehicle status data are acquired. Traffic static data, traffic dynamic data, and vehicle status data are cleaned, spatiotemporally aligned, and standardized to obtain preprocessed traffic static data, traffic dynamic data, and vehicle status data. By using a heterogeneous data fusion algorithm, preprocessed static traffic data, dynamic traffic data, and vehicle status data are semantically correlated and feature-integrated to obtain a vehicle traffic information dataset.

3. The method for unmanned vehicle path planning based on big data according to claim 2, characterized in that, The process of using a heterogeneous data fusion algorithm to semantically associate and integrate preprocessed static traffic data, dynamic traffic data, and vehicle status data to obtain a vehicle traffic information dataset includes the following steps: Establish semantic relationships between static traffic data, dynamic traffic data, and vehicle status data; and extract features from the preprocessed static traffic data, dynamic traffic data, and vehicle status data based on spatial and temporal information to obtain static traffic features, dynamic traffic features, and vehicle status features. Traffic static features, traffic dynamic features, and vehicle state features are standardized; and based on a confidence-weighted fusion strategy, the traffic static features, traffic dynamic features, and vehicle state features are fused to obtain a set of aggregate features. Based on the requirements of path planning, the feature set is structured to generate a vehicle traffic information dataset containing three dimensions: basic road attributes, real-time traffic status, and vehicle dynamic adaptability.

4. The method for unmanned vehicle path planning based on big data according to claim 1, characterized in that, The process of obtaining vehicle route targets and user preferences, and combining them with vehicle traffic information datasets to make route decisions using a pre-configured route decision model to obtain a global initial route includes the following steps: Obtain path goals and user preferences; Based on the path objective, the vehicle traffic information dataset is filtered to obtain the filtered vehicle traffic information; Transform user preferences into path planning constraints; A path decision model is trained based on the improved A* algorithm and reinforcement learning strategy. Using a path decision model, and combining filtered vehicle traffic information and path planning constraints, a path search is performed to obtain a global initial path.

5. The method for unmanned vehicle path planning based on big data according to claim 4, characterized in that, The path decision model trained based on the improved A* algorithm and reinforcement learning strategy includes the following steps: Based on a pre-built historical dataset, combined with an improved A* algorithm, the optimal path sample is obtained; Construct a reinforcement learning model; and train the reinforcement learning model based on the optimal path samples to obtain an initial path decision model; The initial path decision model is iteratively trained using historical datasets, and the hyperparameters of the initial path decision model are optimized through cross-validation to obtain the final path decision model.

6. The method for unmanned vehicle path planning based on big data according to claim 5, characterized in that, The steps of constructing a reinforcement learning model and training it based on optimal path samples to obtain an initial path decision model include: Define the state space, action space, and reward function of the reinforcement learning model; Construct a neural network structure for a reinforcement learning model that includes an input layer, hidden layers, and an output layer; Based on the optimal path samples, the constructed reinforcement learning model is trained to obtain the initial path decision model.

7. The method for unmanned vehicle path planning based on big data according to claim 6, characterized in that, The reward function is expressed as follows: In the formula, R(s,a,s′) represents the immediate reward value obtained when action a transitions from state s to the new state s′; ω1, ω2, ω3, and ω4 represent the weight coefficients of the travel time reward, path distance reward, sample similarity reward, and terminal reward, respectively; R time (s,a,s′) represents the travel time reward; T(s,a,s′) represents the actual travel time from state s to state s′ by performing action a. T ∗ (s,s goal ) represents the improved A* algorithm's computation from state s to the target state s. goal The optimal travel time; α represents the time penalty coefficient, controlling the impact of time deviation on the reward; R dist (s,a,s′) represents the path distance reward; D(s,a,s′) represents the actual path distance from state s to state s′ by performing action a; D ∗ (s,s goal ) represents the state T calculated by the improved A* algorithm. ∗ (from state s to target state s) goal The optimal path distance; β represents the distance penalty coefficient, controlling the degree of influence of distance deviation on the reward; R sim (s,a,s′) represents the sample similarity reward; a * (s) represents the expert recommendation action in state s of the optimal path sample; I(⋅) represents the indicator function; γ represents the similarity reward coefficient, which encourages the model to imitate the decision logic of the optimal sample; R term (s,a,s′) represents the terminal reward; s goal δ represents the target state of the path planning; δ represents the terminal reward coefficient.

8. The method for unmanned vehicle path planning based on big data according to claim 1, characterized in that, The process of continuously acquiring real-time data of the vehicle during the execution of the global initial path, and using dynamic obstacle avoidance algorithms and comprehensive cost functions to dynamically adjust the path until the destination is reached includes the following steps: Continuously acquire real-time data of the vehicle during the execution of the global initial path; By fusing and updating real-time data with vehicle traffic information datasets, a local dynamic environment model for the current moment can be constructed. Based on the global initial path and combined with the local dynamic environment model, the path is adjusted using a dynamic obstacle avoidance algorithm and a comprehensive cost function to obtain the adjusted path and execute it. Continuously acquire real-time data during the execution of the adjusted path, update the local dynamic environment model; and repeatedly adjust the path based on the updated local dynamic environment model, combined with the dynamic obstacle avoidance algorithm and the comprehensive cost function. The system compares the vehicle's current location with the destination coordinates in real time. When the distance between the two is less than a preset threshold, it determines that the destination has been reached and ends the route planning and adjustment process.

9. The method for unmanned vehicle path planning based on big data according to claim 8, characterized in that, The process involves adjusting the path based on a global initial path and a local dynamic environment model using a dynamic obstacle avoidance algorithm to obtain the adjusted path, and then executing the following steps: Based on the global initial path and combined with the local dynamic environment model, several candidate paths are obtained according to the dynamic obstacle avoidance algorithm; Construct a comprehensive cost function that includes safety, comfort, and path tracking performance, and score each candidate path based on the comprehensive cost function; The candidate path with the lowest overall cost is selected as the optimal local path, and the smoothness and feasibility of the optimal local path are verified. The validated optimal local path is used as the adjusted path and executed.

10. A path planning system for unmanned vehicles based on big data, characterized in that, The system includes: The vehicle traffic information dataset construction module is used to acquire static traffic data, dynamic traffic data, and vehicle status data, and then perform data preprocessing and data fusion operations in sequence to obtain the vehicle traffic information dataset. The global initial path construction module is used to obtain vehicle path goals and user preferences, and combined with the vehicle traffic information dataset, it uses a pre-configured path decision model to make path decisions and obtain the global initial path. The real-time local path adjustment module is used to continuously acquire real-time data of the vehicle during the execution of the global initial path, and to dynamically adjust the path using dynamic obstacle avoidance algorithms and comprehensive cost functions until the destination is reached.

Images